<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xml:base="https://tj.ie/" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>TJ Fogarty | Full-Stack AI Engineer</title>
    <link>https://tj.ie/</link>
    <atom:link href="https://tj.ie/index.xml" rel="self" type="application/rss+xml" />
    <description>TJ Fogarty is a full-stack engineer in Ireland building agentic AI for regulated industries, and writing about AI in engineering, the web and making music.</description>
    <language>en</language>
    <item>
      <title>In Two Minds</title>
      <link>https://tj.ie/in-two-minds/</link>
      <description>&lt;p&gt;The idea of a ghost in the machine stemmed from the concept that the mind and body were two separate systems. Today, this doesn’t hold much water. Yet we rebuilt it.&lt;/p&gt;
&lt;p&gt;What I’m interested in is a system with two minds, where one of them belongs to everyone.&lt;/p&gt;
&lt;p&gt;That’s where I feel we are now. I see it with the off shoring of thought to a datacenter residing hundreds of miles away. Your mind posits the question, the thinking machine returns an answer. Maybe it even returns its own questions. The descendant of Google’s &lt;em&gt;“did you mean…?”&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;Where once minds connected, we now share one amongst millions. It feels private, curated. An intimate sharing of information. Knowledge supplies the anchors that keep reality steady.&lt;/p&gt;
&lt;p&gt;You circle deep thought without ever quite going deep enough for fear of running out of breath.&lt;/p&gt;
&lt;p&gt;Except you’re not breathing in the traditional sense. Tokens are your oxygen. Your pockets are your lungs.&lt;/p&gt;
&lt;p&gt;Should you drift further? Do you slip the anchor? You don’t know how you got here, so how can you find your way back? Before you can decide, your thinking mind needs a rest. Give it five hours.&lt;/p&gt;</description>
      <pubDate>Sun, 01 Nov 26 13:06:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/in-two-minds/</guid>
      <content type="html"><![CDATA[
<p>The idea of a ghost in the machine stemmed from the concept that the mind and body were two separate systems. Today, this doesn’t hold much water. Yet we rebuilt it.</p>
<p>What I’m interested in is a system with two minds, where one of them belongs to everyone.</p>
<p>That’s where I feel we are now. I see it with the off shoring of thought to a datacenter residing hundreds of miles away. Your mind posits the question, the thinking machine returns an answer. Maybe it even returns its own questions. The descendant of Google’s <em>“did you mean…?”</em></p>
<p>Where once minds connected, we now share one amongst millions. It feels private, curated. An intimate sharing of information. Knowledge supplies the anchors that keep reality steady.</p>
<p>You circle deep thought without ever quite going deep enough for fear of running out of breath.</p>
<p>Except you’re not breathing in the traditional sense. Tokens are your oxygen. Your pockets are your lungs.</p>
<p>Should you drift further? Do you slip the anchor? You don’t know how you got here, so how can you find your way back? Before you can decide, your thinking mind needs a rest. Give it five hours.</p>
      ]]></content>
    </item>
    <item>
      <title>How Swatcher builds a paint palette</title>
      <link>https://tj.ie/how-swatcher-builds-a-paint-palette/</link>
      <description>&lt;p&gt;Swatcher has had palette generation from the start. Describe a room, get some colours back, matched to real paints. With the &lt;a href=&quot;https://tj.ie/swatcher-v2/&quot;&gt;rebuild&lt;/a&gt; I took the chance to turn it inside out.&lt;/p&gt;
&lt;p&gt;The old way was the obvious one. Ask a model for some hex codes, then find the nearest paint to each. It mostly worked. But nearest can be a long way off, and the model’s reasoning was written for a colour that doesn’t exist. Nobody can walk into a shop and buy &lt;code&gt;#C8B8A2&lt;/code&gt;.&lt;/p&gt;
&lt;p&gt;So now there’s one rule: every colour Swatcher suggests is a real paint. The models never get to make one up. One reads what you wrote, the catalogue does the colour work, and another picks from a short list. That’s really it. Here’s how it hangs together.&lt;/p&gt;
&lt;p&gt;&lt;picture&gt;&lt;source type=&quot;image/webp&quot; srcset=&quot;https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-500.webp 500w, https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-800.webp 800w, https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-1200.webp 1200w&quot; sizes=&quot;(max-width: 730px) 100vw, 730px&quot;&gt;&lt;img loading=&quot;lazy&quot; decoding=&quot;async&quot; src=&quot;https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-500.png&quot; alt=&quot;Flowchart: brief or photo → Reader (Haiku 5.5) → Catalogue resolves → Redirect, Check or Shortlists (up to 8 paints) → Composer → Checked and saved.&quot; width=&quot;1200&quot; height=&quot;594&quot; srcset=&quot;https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-500.png 500w, https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-800.png 800w, https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-1200.png 1200w&quot; sizes=&quot;(max-width: 730px) 100vw, 730px&quot;&gt;&lt;/picture&gt;&lt;/p&gt;
&lt;p&gt;Two model calls, and neither one is ever the source of a colour: the reader only quotes, and the composer can only choose ids from the shortlists.&lt;/p&gt;
&lt;h2 id=&quot;reading-the-brief&quot;&gt;&lt;a href=&quot;#reading-the-brief&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;Reading the Brief&lt;/h2&gt;
&lt;p&gt;The first step is boring on purpose. A small, quick model (Claude Haiku) reads the brief and pulls out what it names. It’s told plainly that it doesn’t choose colours.&lt;/p&gt;
&lt;p&gt;It works out what kind of ask it is: a whole room, one thing like a front door, or just the name of a paint typed into the wrong box. Then it picks out anchors, meaning anything with a colour attached. A paint you want to keep. Oak floors. A red door. “Nothing like Hague Blue.” It also notes which way the room faces, whether you want something a bit lighter than a paint you know, and whether you’ve stuck to one brand.&lt;/p&gt;
&lt;p&gt;What it never does is guess what colour a paint is. It doesn’t know Colourtrend’s Valkriss. The catalogue does.&lt;/p&gt;
&lt;p&gt;Each name goes through the same fuzzy matching as the search box, and it only counts as found if one paint clearly wins. If two paints share a name, you get asked which one you meant. If the whole brief was just a paint’s name, you get sent to that paint’s page instead of a palette of one.&lt;/p&gt;
&lt;p&gt;Photos go the same way. A second reader turns the photo into the same kind of brief, but the colours come from measuring the image’s pixels, not from the model eyeballing it. On a test set of 17 photos that landed closer to the original, and two runs on the same photo agreed about three times as often.&lt;/p&gt;
&lt;h2 id=&quot;shortlists&quot;&gt;&lt;a href=&quot;#shortlists&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;Shortlists&lt;/h2&gt;
&lt;p&gt;This is the bit with no AI in it at all, and it’s where most of the colour work happens. Plain PHP, about 100ms, same answer every time.&lt;/p&gt;
&lt;p&gt;Everything fans out from one starting colour. That’s a paint you said you want, or something already in the room, or the mood the reader picked up. From there it uses the same code behind “what goes with this” on every paint page. That sorts real paints by the job they’d do in a room: the walls, a quieter colour alongside, a white for the woodwork, and an accent.&lt;/p&gt;
&lt;p&gt;I went with jobs rather than colour theory because that’s the question people actually ask. Nobody wants the split-complementary. They want to know which white goes on the skirting.&lt;/p&gt;
&lt;p&gt;Each job gets up to eight paints, then the brief gets applied. A brand limit cuts the catalogue down before anything else, so “only Dulux” can’t leak. Something you’ve ruled out takes its lookalikes from every other brand with it. A paint you already own goes to the top of every list. “A bit lighter” means the same hue a step lighter, and if a range doesn’t have one, it comes back empty rather than handing you something darker.&lt;/p&gt;
&lt;p&gt;One-colour asks get their own shape: the nearest paints, a step lighter and a step darker. An earlier version reached for accents there, and a brown for the hall came back as two greens. Fair enough on paper. Not what was asked for.&lt;/p&gt;
&lt;h2 id=&quot;picking&quot;&gt;&lt;a href=&quot;#picking&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;Picking&lt;/h2&gt;
&lt;p&gt;Now a model gets to have some taste. The composer sees the brief, the fixed colours with their real values, and each job’s shortlist. It picks one paint per job, names the palette, and writes a line about each choice.&lt;/p&gt;
&lt;p&gt;This is the part a lookup can’t do. A north-facing room wants warmer, brighter walls. A red front door should be answered by the accent, not ignored. “Something warm” should get a warm accent, not a cool contrast.&lt;/p&gt;
&lt;p&gt;The trick is in the schema. The only paint IDs it will accept are the ones on this brief’s shortlists, as an enum:&lt;/p&gt;
&lt;pre data-lang=&quot;php&quot; class=&quot;notranslate&quot;&gt;&lt;span class=&quot;hl-value&quot;&gt;&#039;choices&#039;&lt;/span&gt; =&amp;gt; &lt;span class=&quot;hl-variable&quot;&gt;$schema&lt;/span&gt;-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;array&lt;/span&gt;()
    -&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;items&lt;/span&gt;(&lt;span class=&quot;hl-variable&quot;&gt;$schema&lt;/span&gt;-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;object&lt;/span&gt;(&lt;span class=&quot;hl-keyword&quot;&gt;fn&lt;/span&gt; (&lt;span class=&quot;hl-injection&quot;&gt;$schema&lt;/span&gt;) =&amp;gt; [
        &lt;span class=&quot;hl-value&quot;&gt;&#039;role&#039;&lt;/span&gt; =&amp;gt; &lt;span class=&quot;hl-variable&quot;&gt;$schema&lt;/span&gt;-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;string&lt;/span&gt;()-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;enum&lt;/span&gt;(&lt;span class=&quot;hl-property&quot;&gt;array_column&lt;/span&gt;(&lt;span class=&quot;hl-type&quot;&gt;PaletteRole&lt;/span&gt;::&lt;span class=&quot;hl-property&quot;&gt;cases&lt;/span&gt;(), &lt;span class=&quot;hl-value&quot;&gt;&#039;value&#039;&lt;/span&gt;))-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;required&lt;/span&gt;(),
        &lt;span class=&quot;hl-value&quot;&gt;&#039;paint&#039;&lt;/span&gt; =&amp;gt; &lt;span class=&quot;hl-variable&quot;&gt;$schema&lt;/span&gt;-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;string&lt;/span&gt;()-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;enum&lt;/span&gt;(&lt;span class=&quot;hl-variable&quot;&gt;$this&lt;/span&gt;-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;paintIds&lt;/span&gt;)-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;required&lt;/span&gt;(),
        &lt;span class=&quot;hl-value&quot;&gt;&#039;reason&#039;&lt;/span&gt; =&amp;gt; &lt;span class=&quot;hl-variable&quot;&gt;$schema&lt;/span&gt;-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;string&lt;/span&gt;()-&amp;gt;&lt;span class=&quot;hl-property&quot;&gt;required&lt;/span&gt;(),
    ]))
&lt;/pre&gt;
&lt;p&gt;It runs in strict mode, so there’s no other way for it to name a paint. That’s not asking nicely in the prompt. It just can’t produce an ID that isn’t on the list.&lt;/p&gt;
&lt;p&gt;I still check afterwards, since a real ID can land in the wrong job. Anything picked from outside its own list gets dropped, nothing is used twice, and if it forgets a paint you said you own, that goes back in.&lt;/p&gt;
&lt;p&gt;The composer started on Claude Sonnet and now runs on GPT-6 Sol with reasoning off. On 18 real briefs side by side it picked as well, in about 4.5 seconds against 7.4. Thinking didn’t earn its keep here. Once the shortlists had done the narrowing, it didn’t change the picks, only the wait.&lt;/p&gt;
&lt;h2 id=&quot;the-wait&quot;&gt;&lt;a href=&quot;#the-wait&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;The Wait&lt;/h2&gt;
&lt;p&gt;The first version of this took about 20 seconds. That’s a long time to stare at a spinner, and the usual fix is a list of made-up stages on a timer. “Choosing your accent…” People notice.&lt;/p&gt;
&lt;p&gt;The shortlists are ready well before the composer is done, so the form asks for them first. While you wait, you see what it heard: the light, a brand, a paint you named with its colour beside it. Under that is a row of real paints for each job, one lifting at a time. The status only changes when something actually finishes.&lt;/p&gt;
&lt;p&gt;Then I made it faster. Most of the time was the composer thinking. Turning that down got it to about 14 seconds, and the move to Sol brought it under 10. In my last test run, 19 palettes took between 6.5 and 11 seconds, about 9 on average, and photos took about the same as written briefs. You’re looking at shortlists after the first couple of seconds, which helps.&lt;/p&gt;
&lt;h2 id=&quot;the-cost&quot;&gt;&lt;a href=&quot;#the-cost&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;The Cost&lt;/h2&gt;
&lt;p&gt;I ran 19 real generations to see what one actually costs: 12 written briefs and 7 photos, with the token counts logged on every call.&lt;/p&gt;
&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Typical cost&lt;/th&gt;
&lt;th&gt;Range&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;A room scheme from text&lt;/td&gt;
&lt;td&gt;$0.0065&lt;/td&gt;
&lt;td&gt;$0.0061 to $0.0068&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;One colour for one thing&lt;/td&gt;
&lt;td&gt;$0.0040&lt;/td&gt;
&lt;td&gt;$0.0038 to $0.0042&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A scheme from a photo&lt;/td&gt;
&lt;td&gt;$0.0091&lt;/td&gt;
&lt;td&gt;$0.0087 to $0.0101&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;So somewhere between half a cent and a cent. A one-colour ask is cheaper because there are only two jobs to fill. A photo costs more because its colours put more paints on the shortlists, so the composer reads more.&lt;/p&gt;
&lt;p&gt;Almost all of that is the composer: over 90% of every bill. The reader on Haiku costs about $0.0003 for a written brief and $0.0006 for a photo. Sol is priced like Sonnet ($2 per million tokens in, $10 out), so the switch made it faster without making it dearer.&lt;/p&gt;
&lt;p&gt;Keeping the catalogue in charge helps here too. The composer sees eight paints per job, not the thirteen thousand or so in the catalogue, so a prompt stays around 2,000 tokens.&lt;/p&gt;
&lt;h2 id=&quot;what-i-learned&quot;&gt;&lt;a href=&quot;#what-i-learned&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;What I Learned&lt;/h2&gt;
&lt;p&gt;Let the model read and judge, and let the data handle the facts. It’s good at understanding “the door is red and I hate beige”. It’s bad at knowing what colour a specific paint is.&lt;/p&gt;
&lt;p&gt;Make the wrong answer impossible, then check anyway. “Only choose from this list” in a prompt is a request. An enum is a wall.&lt;/p&gt;
&lt;p&gt;Never hand a model an empty list. This one caught me out. A very pale colour has hardly any accents, and “no orange” next to a blue can empty the accent list by itself. Asked to fill a job with nothing in it, the composer made up an ID. Now an empty job gets honest stand-ins, or it’s left out.&lt;/p&gt;
&lt;p&gt;Try things side by side on real briefs before swapping anything. Smaller models were quicker as the reader, and quietly worse. They decided almost everything should be a bit lighter.&lt;/p&gt;
&lt;p&gt;Give it a spin on &lt;a href=&quot;https://swatcher.ie&quot;&gt;Swatcher&lt;/a&gt; and let me know what you think.&lt;/p&gt;</description>
      <pubDate>Fri, 30 Oct 26 14:52:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/how-swatcher-builds-a-paint-palette/</guid>
      <content type="html"><![CDATA[
<p>Swatcher has had palette generation from the start. Describe a room, get some colours back, matched to real paints. With the <a href="https://tj.ie/swatcher-v2/">rebuild</a> I took the chance to turn it inside out.</p>
<p>The old way was the obvious one. Ask a model for some hex codes, then find the nearest paint to each. It mostly worked. But nearest can be a long way off, and the model’s reasoning was written for a colour that doesn’t exist. Nobody can walk into a shop and buy <code>#C8B8A2</code>.</p>
<p>So now there’s one rule: every colour Swatcher suggests is a real paint. The models never get to make one up. One reads what you wrote, the catalogue does the colour work, and another picks from a short list. That’s really it. Here’s how it hangs together.</p>
<p><picture><source type="image/webp" srcset="https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-500.webp 500w, https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-800.webp 800w, https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-1200.webp 1200w" sizes="(max-width: 730px) 100vw, 730px"><img loading="lazy" decoding="async" src="https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-500.png" alt="Flowchart: brief or photo → Reader (Haiku 5.5) → Catalogue resolves → Redirect, Check or Shortlists (up to 8 paints) → Composer → Checked and saved." width="1200" height="594" srcset="https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-500.png 500w, https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-800.png 800w, https://fls-a30f6847-d5e0-4bf4-9d5e-bc7b9e178222.laravel.cloud/palette-pipeline-7ixweu-1200.png 1200w" sizes="(max-width: 730px) 100vw, 730px"></picture></p>
<p>Two model calls, and neither one is ever the source of a colour: the reader only quotes, and the composer can only choose ids from the shortlists.</p>
<h2 id="reading-the-brief"><a href="#reading-the-brief" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>Reading the Brief</h2>
<p>The first step is boring on purpose. A small, quick model (Claude Haiku) reads the brief and pulls out what it names. It’s told plainly that it doesn’t choose colours.</p>
<p>It works out what kind of ask it is: a whole room, one thing like a front door, or just the name of a paint typed into the wrong box. Then it picks out anchors, meaning anything with a colour attached. A paint you want to keep. Oak floors. A red door. “Nothing like Hague Blue.” It also notes which way the room faces, whether you want something a bit lighter than a paint you know, and whether you’ve stuck to one brand.</p>
<p>What it never does is guess what colour a paint is. It doesn’t know Colourtrend’s Valkriss. The catalogue does.</p>
<p>Each name goes through the same fuzzy matching as the search box, and it only counts as found if one paint clearly wins. If two paints share a name, you get asked which one you meant. If the whole brief was just a paint’s name, you get sent to that paint’s page instead of a palette of one.</p>
<p>Photos go the same way. A second reader turns the photo into the same kind of brief, but the colours come from measuring the image’s pixels, not from the model eyeballing it. On a test set of 17 photos that landed closer to the original, and two runs on the same photo agreed about three times as often.</p>
<h2 id="shortlists"><a href="#shortlists" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>Shortlists</h2>
<p>This is the bit with no AI in it at all, and it’s where most of the colour work happens. Plain PHP, about 100ms, same answer every time.</p>
<p>Everything fans out from one starting colour. That’s a paint you said you want, or something already in the room, or the mood the reader picked up. From there it uses the same code behind “what goes with this” on every paint page. That sorts real paints by the job they’d do in a room: the walls, a quieter colour alongside, a white for the woodwork, and an accent.</p>
<p>I went with jobs rather than colour theory because that’s the question people actually ask. Nobody wants the split-complementary. They want to know which white goes on the skirting.</p>
<p>Each job gets up to eight paints, then the brief gets applied. A brand limit cuts the catalogue down before anything else, so “only Dulux” can’t leak. Something you’ve ruled out takes its lookalikes from every other brand with it. A paint you already own goes to the top of every list. “A bit lighter” means the same hue a step lighter, and if a range doesn’t have one, it comes back empty rather than handing you something darker.</p>
<p>One-colour asks get their own shape: the nearest paints, a step lighter and a step darker. An earlier version reached for accents there, and a brown for the hall came back as two greens. Fair enough on paper. Not what was asked for.</p>
<h2 id="picking"><a href="#picking" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>Picking</h2>
<p>Now a model gets to have some taste. The composer sees the brief, the fixed colours with their real values, and each job’s shortlist. It picks one paint per job, names the palette, and writes a line about each choice.</p>
<p>This is the part a lookup can’t do. A north-facing room wants warmer, brighter walls. A red front door should be answered by the accent, not ignored. “Something warm” should get a warm accent, not a cool contrast.</p>
<p>The trick is in the schema. The only paint IDs it will accept are the ones on this brief’s shortlists, as an enum:</p>
<pre data-lang="php" class="notranslate"><span class="hl-value">'choices'</span> =&gt; <span class="hl-variable">$schema</span>-&gt;<span class="hl-property">array</span>()
    -&gt;<span class="hl-property">items</span>(<span class="hl-variable">$schema</span>-&gt;<span class="hl-property">object</span>(<span class="hl-keyword">fn</span> (<span class="hl-injection">$schema</span>) =&gt; [
        <span class="hl-value">'role'</span> =&gt; <span class="hl-variable">$schema</span>-&gt;<span class="hl-property">string</span>()-&gt;<span class="hl-property">enum</span>(<span class="hl-property">array_column</span>(<span class="hl-type">PaletteRole</span>::<span class="hl-property">cases</span>(), <span class="hl-value">'value'</span>))-&gt;<span class="hl-property">required</span>(),
        <span class="hl-value">'paint'</span> =&gt; <span class="hl-variable">$schema</span>-&gt;<span class="hl-property">string</span>()-&gt;<span class="hl-property">enum</span>(<span class="hl-variable">$this</span>-&gt;<span class="hl-property">paintIds</span>)-&gt;<span class="hl-property">required</span>(),
        <span class="hl-value">'reason'</span> =&gt; <span class="hl-variable">$schema</span>-&gt;<span class="hl-property">string</span>()-&gt;<span class="hl-property">required</span>(),
    ]))
</pre>
<p>It runs in strict mode, so there’s no other way for it to name a paint. That’s not asking nicely in the prompt. It just can’t produce an ID that isn’t on the list.</p>
<p>I still check afterwards, since a real ID can land in the wrong job. Anything picked from outside its own list gets dropped, nothing is used twice, and if it forgets a paint you said you own, that goes back in.</p>
<p>The composer started on Claude Sonnet and now runs on GPT-6 Sol with reasoning off. On 18 real briefs side by side it picked as well, in about 4.5 seconds against 7.4. Thinking didn’t earn its keep here. Once the shortlists had done the narrowing, it didn’t change the picks, only the wait.</p>
<h2 id="the-wait"><a href="#the-wait" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>The Wait</h2>
<p>The first version of this took about 20 seconds. That’s a long time to stare at a spinner, and the usual fix is a list of made-up stages on a timer. “Choosing your accent…” People notice.</p>
<p>The shortlists are ready well before the composer is done, so the form asks for them first. While you wait, you see what it heard: the light, a brand, a paint you named with its colour beside it. Under that is a row of real paints for each job, one lifting at a time. The status only changes when something actually finishes.</p>
<p>Then I made it faster. Most of the time was the composer thinking. Turning that down got it to about 14 seconds, and the move to Sol brought it under 10. In my last test run, 19 palettes took between 6.5 and 11 seconds, about 9 on average, and photos took about the same as written briefs. You’re looking at shortlists after the first couple of seconds, which helps.</p>
<h2 id="the-cost"><a href="#the-cost" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>The Cost</h2>
<p>I ran 19 real generations to see what one actually costs: 12 written briefs and 7 photos, with the token counts logged on every call.</p>
<table>
<thead>
<tr>
<th></th>
<th>Typical cost</th>
<th>Range</th>
</tr>
</thead>
<tbody>
<tr>
<td>A room scheme from text</td>
<td>$0.0065</td>
<td>$0.0061 to $0.0068</td>
</tr>
<tr>
<td>One colour for one thing</td>
<td>$0.0040</td>
<td>$0.0038 to $0.0042</td>
</tr>
<tr>
<td>A scheme from a photo</td>
<td>$0.0091</td>
<td>$0.0087 to $0.0101</td>
</tr>
</tbody>
</table>
<p>So somewhere between half a cent and a cent. A one-colour ask is cheaper because there are only two jobs to fill. A photo costs more because its colours put more paints on the shortlists, so the composer reads more.</p>
<p>Almost all of that is the composer: over 90% of every bill. The reader on Haiku costs about $0.0003 for a written brief and $0.0006 for a photo. Sol is priced like Sonnet ($2 per million tokens in, $10 out), so the switch made it faster without making it dearer.</p>
<p>Keeping the catalogue in charge helps here too. The composer sees eight paints per job, not the thirteen thousand or so in the catalogue, so a prompt stays around 2,000 tokens.</p>
<h2 id="what-i-learned"><a href="#what-i-learned" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>What I Learned</h2>
<p>Let the model read and judge, and let the data handle the facts. It’s good at understanding “the door is red and I hate beige”. It’s bad at knowing what colour a specific paint is.</p>
<p>Make the wrong answer impossible, then check anyway. “Only choose from this list” in a prompt is a request. An enum is a wall.</p>
<p>Never hand a model an empty list. This one caught me out. A very pale colour has hardly any accents, and “no orange” next to a blue can empty the accent list by itself. Asked to fill a job with nothing in it, the composer made up an ID. Now an empty job gets honest stand-ins, or it’s left out.</p>
<p>Try things side by side on real briefs before swapping anything. Smaller models were quicker as the reader, and quietly worse. They decided almost everything should be a bit lighter.</p>
<p>Give it a spin on <a href="https://swatcher.ie">Swatcher</a> and let me know what you think.</p>
      ]]></content>
    </item>
    <item>
      <title>Way Back When</title>
      <link>https://tj.ie/way-back-when/</link>
      <description>&lt;p&gt;What we have today is an unimaginable leap in capability and features when it comes to development.&lt;/p&gt;
&lt;p&gt;Instead of tracing AI psychosis, I thought it would be fun to catalogue a few things that were normal back then. It might be funny. I couldn’t recall everything anyway. My mind probably won’t let me. For safety reasons.&lt;/p&gt;
&lt;h2 id=&quot;unfettered-internet-access&quot;&gt;&lt;a href=&quot;#unfettered-internet-access&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;Unfettered Internet Access&lt;/h2&gt;
&lt;p&gt;My first memory of contributing to the web was in the early 2000s, on phpBB forums. There were these short codes, or BBCodes as they were officially called. They were mostly used for formatting, like &lt;code&gt;[i]italic[/i]&lt;/code&gt;, but there might have been custom ones for different functions as well. I can’t remember exactly. Maybe my mind is telling me something there.&lt;/p&gt;
&lt;p&gt;I made “signatures” for myself and others: little 300x150 pixel graphics, emblazoned with a username. I’d boot up a cracked version of Photoshop, create clouds, layer brushes, warp lens flares, and slap on a “render” (a cutout) of a video game character. You’d bang out a few of those a week for friends or folks on the forum.&lt;/p&gt;
&lt;h2 id=&quot;a-more-deliberate-direction&quot;&gt;&lt;a href=&quot;#a-more-deliberate-direction&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;A More Deliberate Direction&lt;/h2&gt;
&lt;p&gt;In college I learned how to build websites and write software. This involved Flash, HTML and Java.&lt;/p&gt;
&lt;p&gt;The Flash approach was more interesting. Less limiting. We wrote ActionScript and designed custom graphics. We’ve been striving to return to that way for a while, but it’s not the same.&lt;/p&gt;
&lt;p&gt;We would create websites in Photoshop (a legit copy). From there we’d slice up the image in Dreamweaver and jam those images into tables. Responsive design wasn’t a thing yet. Or accessibility, really. Though in fairness, you weren’t limited to CSS; it was whatever you could imagine.&lt;/p&gt;
&lt;p&gt;Outside of that there was Java, with NetBeans and Eclipse. We had a Mac lab that ran Parallels so we could use Windows. It felt like a waste of decent hardware, but it gave me some exposure to Mac, even if it was just booting up and launching a program.&lt;/p&gt;
&lt;p&gt;I remember we had an Android Development module, and I got a talking-to for writing my app in a webview. It was still technically an app that fulfilled the brief, but it wasn’t in the spirit of things. Sorry, Ita.&lt;/p&gt;
&lt;h2 id=&quot;first-blood&quot;&gt;&lt;a href=&quot;#first-blood&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;First Blood&lt;/h2&gt;
&lt;p&gt;I did a work placement as part of my course, and it was mostly graphic design and front-end fixes. Most of these fixes were for Internet Explorer 6.&lt;/p&gt;
&lt;p&gt;Quirks mode. Padding. Spacer GIFs. &lt;code&gt;zoom: 1&lt;/code&gt;. It was wild, and all of it was just to make sure our changes didn’t break anything in IE7. We monitored any chatter about support for IE6 being dropped, but that didn’t help when customers were still using it.&lt;/p&gt;
&lt;h2 id=&quot;fake-it-til-you-make-it&quot;&gt;&lt;a href=&quot;#fake-it-til-you-make-it&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;Fake It ’Til You Make It&lt;/h2&gt;
&lt;p&gt;I graduated and landed a job as a web developer. I remember being shown fancy-looking buttons with gradients in the interview, and I assured them I could do that. I didn’t say I could do it &lt;em&gt;right now&lt;/em&gt;, but I was sure I could figure it out.&lt;/p&gt;
&lt;p&gt;I brought along my knowledge of IE6. It felt like that wasn’t going anywhere anytime soon.&lt;/p&gt;
&lt;p&gt;Designers loved gradients, but some browsers didn’t. That’s where the likes of CSS3 PIE came in. Drop in a &lt;code&gt;.htc&lt;/code&gt; file and everyone was happy.&lt;/p&gt;
&lt;p&gt;That’s where I met WordPress. We had been doing static websites, but now customers wanted more control, rather than emailing us with copy updates. Understandable.&lt;/p&gt;
&lt;p&gt;WordPress led to Joomla! and Drupal. I had no idea about them, but figured it out along the way. We even had a customer request a WordPress installation inside of their Joomla! installation. It worked, but felt wrong. It was like asking a sparrow to lay a goose egg. You just make the swap when no one’s looking and exclaim &lt;em&gt;TA-DA&lt;/em&gt;!&lt;/p&gt;
&lt;h2 id=&quot;what-starts-well&quot;&gt;&lt;a href=&quot;#what-starts-well&quot; class=&quot;c-post__direct-link&quot; aria-hidden=&quot;true&quot; tabindex=&quot;-1&quot; title=&quot;Jump to heading&quot;&gt;#&lt;/a&gt;What Starts Well&lt;/h2&gt;
&lt;p&gt;There were moments of pure creativity built from necessity, which was the most exciting part. Could I compare that to the landscape of today? Not so much, I think. We have the tools now in abundance, all thanks to those wild years of figuring things out.&lt;/p&gt;</description>
      <pubDate>Tue, 27 Oct 26 15:20:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/way-back-when/</guid>
      <content type="html"><![CDATA[
<p>What we have today is an unimaginable leap in capability and features when it comes to development.</p>
<p>Instead of tracing AI psychosis, I thought it would be fun to catalogue a few things that were normal back then. It might be funny. I couldn’t recall everything anyway. My mind probably won’t let me. For safety reasons.</p>
<h2 id="unfettered-internet-access"><a href="#unfettered-internet-access" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>Unfettered Internet Access</h2>
<p>My first memory of contributing to the web was in the early 2000s, on phpBB forums. There were these short codes, or BBCodes as they were officially called. They were mostly used for formatting, like <code>[i]italic[/i]</code>, but there might have been custom ones for different functions as well. I can’t remember exactly. Maybe my mind is telling me something there.</p>
<p>I made “signatures” for myself and others: little 300x150 pixel graphics, emblazoned with a username. I’d boot up a cracked version of Photoshop, create clouds, layer brushes, warp lens flares, and slap on a “render” (a cutout) of a video game character. You’d bang out a few of those a week for friends or folks on the forum.</p>
<h2 id="a-more-deliberate-direction"><a href="#a-more-deliberate-direction" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>A More Deliberate Direction</h2>
<p>In college I learned how to build websites and write software. This involved Flash, HTML and Java.</p>
<p>The Flash approach was more interesting. Less limiting. We wrote ActionScript and designed custom graphics. We’ve been striving to return to that way for a while, but it’s not the same.</p>
<p>We would create websites in Photoshop (a legit copy). From there we’d slice up the image in Dreamweaver and jam those images into tables. Responsive design wasn’t a thing yet. Or accessibility, really. Though in fairness, you weren’t limited to CSS; it was whatever you could imagine.</p>
<p>Outside of that there was Java, with NetBeans and Eclipse. We had a Mac lab that ran Parallels so we could use Windows. It felt like a waste of decent hardware, but it gave me some exposure to Mac, even if it was just booting up and launching a program.</p>
<p>I remember we had an Android Development module, and I got a talking-to for writing my app in a webview. It was still technically an app that fulfilled the brief, but it wasn’t in the spirit of things. Sorry, Ita.</p>
<h2 id="first-blood"><a href="#first-blood" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>First Blood</h2>
<p>I did a work placement as part of my course, and it was mostly graphic design and front-end fixes. Most of these fixes were for Internet Explorer 6.</p>
<p>Quirks mode. Padding. Spacer GIFs. <code>zoom: 1</code>. It was wild, and all of it was just to make sure our changes didn’t break anything in IE7. We monitored any chatter about support for IE6 being dropped, but that didn’t help when customers were still using it.</p>
<h2 id="fake-it-til-you-make-it"><a href="#fake-it-til-you-make-it" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>Fake It ’Til You Make It</h2>
<p>I graduated and landed a job as a web developer. I remember being shown fancy-looking buttons with gradients in the interview, and I assured them I could do that. I didn’t say I could do it <em>right now</em>, but I was sure I could figure it out.</p>
<p>I brought along my knowledge of IE6. It felt like that wasn’t going anywhere anytime soon.</p>
<p>Designers loved gradients, but some browsers didn’t. That’s where the likes of CSS3 PIE came in. Drop in a <code>.htc</code> file and everyone was happy.</p>
<p>That’s where I met WordPress. We had been doing static websites, but now customers wanted more control, rather than emailing us with copy updates. Understandable.</p>
<p>WordPress led to Joomla! and Drupal. I had no idea about them, but figured it out along the way. We even had a customer request a WordPress installation inside of their Joomla! installation. It worked, but felt wrong. It was like asking a sparrow to lay a goose egg. You just make the swap when no one’s looking and exclaim <em>TA-DA</em>!</p>
<h2 id="what-starts-well"><a href="#what-starts-well" class="c-post__direct-link" aria-hidden="true" tabindex="-1" title="Jump to heading">#</a>What Starts Well</h2>
<p>There were moments of pure creativity built from necessity, which was the most exciting part. Could I compare that to the landscape of today? Not so much, I think. We have the tools now in abundance, all thanks to those wild years of figuring things out.</p>
      ]]></content>
    </item>
    <item>
      <title>I&#039;ve Been Here Before</title>
      <link>https://tj.ie/migrating-my-site-to-custom-cms/</link>
      <description>&lt;p&gt;This site has been rebuilt from the inside, and if I did it right, you won’t notice. Same design. Same URLs. Same feed.&lt;/p&gt;
&lt;p&gt;It runs on Laravel now, with an editor that lets me write, stash drafts and come back to them without fighting the system. It’s something the old setup couldn’t do. I’m a developer who gets impatient with his own writing, so anything that removes friction is a net win.&lt;/p&gt;
&lt;p&gt;The cover is still a fitted sheet. It stretches over whatever’s underneath and nobody sees the corners. I’ve done migrations like this by hand more times than I’d like to admit. This time I asked AI, and watched.&lt;/p&gt;
&lt;p&gt;I know the shape of the thing better than I know every seam. Somewhere in the new walls there are little ghosts I haven’t met yet, or maybe I’m the ghost in the machine now.&lt;/p&gt;
&lt;p&gt;It looks the same, and it’ll change again. I know that much.&lt;/p&gt;</description>
      <pubDate>Sun, 25 Oct 26 08:59:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/migrating-my-site-to-custom-cms/</guid>
      <content type="html"><![CDATA[
<p>This site has been rebuilt from the inside, and if I did it right, you won’t notice. Same design. Same URLs. Same feed.</p>
<p>It runs on Laravel now, with an editor that lets me write, stash drafts and come back to them without fighting the system. It’s something the old setup couldn’t do. I’m a developer who gets impatient with his own writing, so anything that removes friction is a net win.</p>
<p>The cover is still a fitted sheet. It stretches over whatever’s underneath and nobody sees the corners. I’ve done migrations like this by hand more times than I’d like to admit. This time I asked AI, and watched.</p>
<p>I know the shape of the thing better than I know every seam. Somewhere in the new walls there are little ghosts I haven’t met yet, or maybe I’m the ghost in the machine now.</p>
<p>It looks the same, and it’ll change again. I know that much.</p>
      ]]></content>
    </item>
    <item>
      <title>Swatcher 2.0</title>
      <link>https://tj.ie/swatcher-v2/</link>
      <description>&lt;p&gt;When &lt;a href=&quot;https://swatcher.ie&quot;&gt;Swatcher&lt;/a&gt; was first released, it was a fairly simple concept - pull together paints from different brands available in Ireland, and explore them from one place. Add some colour generation from prompts and match them to real paints.&lt;/p&gt;&lt;p&gt;I wasn’t expecting it to take off, but it did. It was both exciting and nerve-wracking.&lt;/p&gt;&lt;p&gt;It was strung together with tape, connecting different parts. There was Firebase for authentication and storing user data, Netlify for hosting, with serverless functions to handle things like server-side rendering for SEO, and JSON files of paint data that had to be manually updated and pushed live.&lt;/p&gt;&lt;p&gt;I thought I was doing something useful by breaking out the paint data into its own separate API. Now the paint data had to be shared between that and the front-end to generate the sitemaps. I did this to myself.&lt;/p&gt;&lt;p&gt;Fun started to become a chore, and I wasn’t enjoying being in maintenance mode when I could have been exploring more ideas and iterating on something without fear of it crumbling.&lt;/p&gt;&lt;hr&gt;&lt;p&gt;Over the last little while, I’ve been working on a rebuild. Something a bit more sturdy and future-proof.&lt;/p&gt;&lt;p&gt;Last year I built an app for my wife and used Laravel hosted on Laravel Cloud. It’s been easy enough to keep going and improving thanks to everything living in one place. Plus tests. Lots of tests. The luxury of working with a real database and migrations meant I wasn’t worrying about everything falling apart with one misstep.&lt;/p&gt;&lt;p&gt;So I took the same approach here. Swatcher 2.0 is now live and has a bunch of affordances baked in. The big one being I have an admin dashboard to upload paint data, which gets persisted to the database. No more placing files in the right place with the right name. It’s a quick job that I don’t mind doing now.&lt;/p&gt;&lt;p&gt;I’m still using Vue on the front-end, but with Inertia and SSR for SEO.&lt;/p&gt;&lt;p&gt;Give it a spin and let me know what you think.&lt;/p&gt;</description>
      <pubDate>Thu, 03 Sep 26 09:00:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/swatcher-v2/</guid>
      <content type="html"><![CDATA[
<p>When <a href="https://swatcher.ie">Swatcher</a> was first released, it was a fairly simple concept - pull together paints from different brands available in Ireland, and explore them from one place. Add some colour generation from prompts and match them to real paints.</p><p>I wasn’t expecting it to take off, but it did. It was both exciting and nerve-wracking.</p><p>It was strung together with tape, connecting different parts. There was Firebase for authentication and storing user data, Netlify for hosting, with serverless functions to handle things like server-side rendering for SEO, and JSON files of paint data that had to be manually updated and pushed live.</p><p>I thought I was doing something useful by breaking out the paint data into its own separate API. Now the paint data had to be shared between that and the front-end to generate the sitemaps. I did this to myself.</p><p>Fun started to become a chore, and I wasn’t enjoying being in maintenance mode when I could have been exploring more ideas and iterating on something without fear of it crumbling.</p><hr><p>Over the last little while, I’ve been working on a rebuild. Something a bit more sturdy and future-proof.</p><p>Last year I built an app for my wife and used Laravel hosted on Laravel Cloud. It’s been easy enough to keep going and improving thanks to everything living in one place. Plus tests. Lots of tests. The luxury of working with a real database and migrations meant I wasn’t worrying about everything falling apart with one misstep.</p><p>So I took the same approach here. Swatcher 2.0 is now live and has a bunch of affordances baked in. The big one being I have an admin dashboard to upload paint data, which gets persisted to the database. No more placing files in the right place with the right name. It’s a quick job that I don’t mind doing now.</p><p>I’m still using Vue on the front-end, but with Inertia and SSR for SEO.</p><p>Give it a spin and let me know what you think.</p>
      ]]></content>
    </item>
    <item>
      <title>A Collection of Music Tools</title>
      <link>https://tj.ie/music-tools/</link>
      <description>&lt;p&gt;I’ve been getting more into making music lately, especially around areas of writing interesting parts and how to mix them well. It’s a steep learning curve if you want to go beyond hitting record and leaving it at that, so I used some free credits to generate some tools to help out.&lt;/p&gt;&lt;p&gt;It’s completely AI coded, and might not have all the right information there, but I guess I’ll learn that lesson the hard way when it comes to it. Whatever about having to make the next big thing with AI or be jobless forever, I’m have some fun spinning up things useful to me. Maybe they’ll be useful to you, too.&lt;/p&gt;&lt;p&gt;You can see what’s available at &lt;a href=&quot;https://music.tj.ie&quot;&gt;music.tj.ie&lt;/a&gt;&lt;/p&gt;</description>
      <pubDate>Sat, 25 Jul 26 20:22:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/music-tools/</guid>
      <content type="html"><![CDATA[
<p>I’ve been getting more into making music lately, especially around areas of writing interesting parts and how to mix them well. It’s a steep learning curve if you want to go beyond hitting record and leaving it at that, so I used some free credits to generate some tools to help out.</p><p>It’s completely AI coded, and might not have all the right information there, but I guess I’ll learn that lesson the hard way when it comes to it. Whatever about having to make the next big thing with AI or be jobless forever, I’m have some fun spinning up things useful to me. Maybe they’ll be useful to you, too.</p><p>You can see what’s available at <a href="https://music.tj.ie">music.tj.ie</a></p>
      ]]></content>
    </item>
    <item>
      <title>Making More Music</title>
      <link>https://tj.ie/making-more-music/</link>
      <description>&lt;p&gt;One of the fun parts when learning to make music is playing around with different sounds to see what it kicks off. Pianos and cellos often evoke a cinematic sort of mood in me, so after coming up with a simple piano melody, I layered it over time, then stripped it back. The melody stuck in my head for days, and I think I’ll expand on it in the future with a “main theme” type approach.&lt;/p&gt;&lt;p&gt;I’ll probably go back to something guitar-driven next. I have some new guitar strings on the way which will let me tune a bit lower. I’ll have to break the 7-string guitar out for something at some stage.&lt;/p&gt;&lt;iframe data-testid=&quot;embed-iframe&quot; style=&quot;border-radius:12px&quot; src=&quot;https://open.spotify.com/embed/track/7lqs1RihztNzs6V5CiHQ4d?utm_source=generator&amp;theme=0&quot; width=&quot;100%&quot; height=&quot;352&quot; frameborder=&quot;0&quot; allowfullscreen=&quot;&quot; allow=&quot;autoplay; clipboard-write; encrypted-media; fullscreen; picture-in-picture&quot; loading=&quot;lazy&quot;&gt;&lt;/iframe&gt;</description>
      <pubDate>Sun, 10 May 26 09:00:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/making-more-music/</guid>
      <content type="html"><![CDATA[
<p>One of the fun parts when learning to make music is playing around with different sounds to see what it kicks off. Pianos and cellos often evoke a cinematic sort of mood in me, so after coming up with a simple piano melody, I layered it over time, then stripped it back. The melody stuck in my head for days, and I think I’ll expand on it in the future with a “main theme” type approach.</p><p>I’ll probably go back to something guitar-driven next. I have some new guitar strings on the way which will let me tune a bit lower. I’ll have to break the 7-string guitar out for something at some stage.</p><iframe data-testid="embed-iframe" style="border-radius:12px" src="https://open.spotify.com/embed/track/7lqs1RihztNzs6V5CiHQ4d?utm_source=generator&theme=0" width="100%" height="352" frameborder="0" allowfullscreen="" allow="autoplay; clipboard-write; encrypted-media; fullscreen; picture-in-picture" loading="lazy"></iframe>
      ]]></content>
    </item>
    <item>
      <title>Hey, I wrote a song</title>
      <link>https://tj.ie/making-music/</link>
      <description>&lt;p&gt;I’ve been toying with the idea of writing music for many years now. I tried it nearly 20 years ago, directly wiring up my amp to my desktop computer using software called Mixcraft. It was version 3 at the time, I think.&lt;/p&gt;&lt;p&gt;It didn’t really pan out. I just stuck to recording guitar covers, with a few original recordings here and there, but nothing substantial.&lt;/p&gt;&lt;hr&gt;&lt;p&gt;Over the last year, I’ve really challenged myself to learn the software and invest in myself to help me enjoy the process of making music. I’ve had a domain - &lt;a href=&quot;https://hauntedlogic.com&quot;&gt;hauntedlogic.com&lt;/a&gt; for the better part of a decade. I just liked the name, but didn’t know what to use it for. Now I do.&lt;/p&gt;&lt;p&gt;With that, here’s the first song I’ve ever written. It’s not perfect, but it was a lot of fun, and I feel pretty happy for sticking with it.&lt;/p&gt;&lt;iframe width=&quot;100%&quot; height=&quot;166&quot; scrolling=&quot;no&quot; frameborder=&quot;no&quot; allow=&quot;autoplay&quot; src=&quot;https://w.soundcloud.com/player/?url=https%3A//api.soundcloud.com/tracks/soundcloud%253Atracks%253A2313384728&amp;color=%230c0c0a&amp;auto_play=false&amp;hide_related=false&amp;show_comments=true&amp;show_user=true&amp;show_reposts=false&amp;show_teaser=true&quot;&gt;&lt;/iframe&gt;&lt;div style=&quot;font-size: 10px; color: #cccccc;line-break: anywhere;word-break: normal;overflow: hidden;white-space: nowrap;text-overflow: ellipsis; font-family: Interstate,Lucida Grande,Lucida Sans Unicode,Lucida Sans,Garuda,Verdana,Tahoma,sans-serif;font-weight: 100;&quot;&gt;&lt;a href=&quot;https://soundcloud.com/tjfogarty&quot; title=&quot;tjfogarty&quot; target=&quot;_blank&quot; style=&quot;color: #cccccc; text-decoration: none;&quot;&gt;tjfogarty&lt;/a&gt; · &lt;a href=&quot;https://soundcloud.com/tjfogarty/lift&quot; title=&quot;lift&quot; target=&quot;_blank&quot; style=&quot;color: #cccccc; text-decoration: none;&quot;&gt;lift&lt;/a&gt;&lt;/div&gt;&lt;iframe data-testid=&quot;embed-iframe&quot; style=&quot;border-radius:12px;margin-top:30px&quot; src=&quot;https://open.spotify.com/embed/track/22ZjpFFbw9DQ6aUkoaCBeE?utm_source=generator&amp;theme=0&quot; width=&quot;100%&quot; height=&quot;352&quot; frameborder=&quot;0&quot; allowfullscreen=&quot;&quot; allow=&quot;autoplay; clipboard-write; encrypted-media; fullscreen; picture-in-picture&quot; loading=&quot;lazy&quot;&gt;&lt;/iframe&gt;</description>
      <pubDate>Sat, 02 May 26 09:00:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/making-music/</guid>
      <content type="html"><![CDATA[
<p>I’ve been toying with the idea of writing music for many years now. I tried it nearly 20 years ago, directly wiring up my amp to my desktop computer using software called Mixcraft. It was version 3 at the time, I think.</p><p>It didn’t really pan out. I just stuck to recording guitar covers, with a few original recordings here and there, but nothing substantial.</p><hr><p>Over the last year, I’ve really challenged myself to learn the software and invest in myself to help me enjoy the process of making music. I’ve had a domain - <a href="https://hauntedlogic.com">hauntedlogic.com</a> for the better part of a decade. I just liked the name, but didn’t know what to use it for. Now I do.</p><p>With that, here’s the first song I’ve ever written. It’s not perfect, but it was a lot of fun, and I feel pretty happy for sticking with it.</p><iframe width="100%" height="166" scrolling="no" frameborder="no" allow="autoplay" src="https://w.soundcloud.com/player/?url=https%3A//api.soundcloud.com/tracks/soundcloud%253Atracks%253A2313384728&color=%230c0c0a&auto_play=false&hide_related=false&show_comments=true&show_user=true&show_reposts=false&show_teaser=true"></iframe><div style="font-size: 10px; color: #cccccc;line-break: anywhere;word-break: normal;overflow: hidden;white-space: nowrap;text-overflow: ellipsis; font-family: Interstate,Lucida Grande,Lucida Sans Unicode,Lucida Sans,Garuda,Verdana,Tahoma,sans-serif;font-weight: 100;"><a href="https://soundcloud.com/tjfogarty" title="tjfogarty" target="_blank" style="color: #cccccc; text-decoration: none;">tjfogarty</a> · <a href="https://soundcloud.com/tjfogarty/lift" title="lift" target="_blank" style="color: #cccccc; text-decoration: none;">lift</a></div><iframe data-testid="embed-iframe" style="border-radius:12px;margin-top:30px" src="https://open.spotify.com/embed/track/22ZjpFFbw9DQ6aUkoaCBeE?utm_source=generator&theme=0" width="100%" height="352" frameborder="0" allowfullscreen="" allow="autoplay; clipboard-write; encrypted-media; fullscreen; picture-in-picture" loading="lazy"></iframe>
      ]]></content>
    </item>
    <item>
      <title>What&#039;s next for Swatcher</title>
      <link>https://tj.ie/whats-next-for-swatcher/</link>
      <description>&lt;p&gt;&lt;a href=&quot;https://swatcher.ie&quot;&gt;Swatcher&lt;/a&gt; was officially launched in May of 2025, and in that time, it has gone through many updates. It has exceeded my own expectations in terms of traction and gets over a thousand visitors a month.&lt;/p&gt;&lt;p&gt;The vast majority of those users come from Google, and they land on a specific paint detail page. From there, they typically visit the paint page on the brand’s website.&lt;/p&gt;&lt;p&gt;That was my intended goal as a value-first approach. I wasn’t happy with having to jump around different sites to compare colours, so I wanted a system where I could view a colour and see suggested complementary colours.&lt;/p&gt;&lt;p&gt;There’s also the Text to Paint feature. You give it a description, and it generates a palette of paint colours for you to browse. It’s been used nearly 900 times in 6 months, which is pretty cool to see.&lt;/p&gt;&lt;p&gt;It was only recently that I took a punt on implementing Stripe to introduce a “Pro” plan. Features aren’t gated, but limits are imposed as they cost money to run. Again, keeping it value-first. The main reason people are there is to browse paint colours, and you don’t need an account to do that.&lt;/p&gt;&lt;p&gt;What’s next?&lt;/p&gt;&lt;p&gt;Well, I don’t see Swatcher changing too much. It’ll still be maintained and updated as it has been. Features may be slower to materialise, but only because it already does what it needs to. I could do a better job of surfacing different features, like pointing your camera at something to get a suggested matching paint colour, but I don’t see any massive features coming soon.&lt;/p&gt;&lt;p&gt;It was architected in a way that the API is independent of the Swatcher front-end. I did this because I was reaching out to paint brands, and could suggest being able to spin up a brand-specific site to showcase their paint colours. I didn’t get any responses, so now I’m thinking about it differently.&lt;/p&gt;&lt;p&gt;I have a few ideas that I’m working on that can be built from the Swatcher API. I’m pretty excited to explore this, and I’ll post updates if it goes anywhere. Otherwise, you won’t hear from me for a good few months.&lt;/p&gt;</description>
      <pubDate>Fri, 20 Feb 26 09:00:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/whats-next-for-swatcher/</guid>
      <content type="html"><![CDATA[
<p><a href="https://swatcher.ie">Swatcher</a> was officially launched in May of 2025, and in that time, it has gone through many updates. It has exceeded my own expectations in terms of traction and gets over a thousand visitors a month.</p><p>The vast majority of those users come from Google, and they land on a specific paint detail page. From there, they typically visit the paint page on the brand’s website.</p><p>That was my intended goal as a value-first approach. I wasn’t happy with having to jump around different sites to compare colours, so I wanted a system where I could view a colour and see suggested complementary colours.</p><p>There’s also the Text to Paint feature. You give it a description, and it generates a palette of paint colours for you to browse. It’s been used nearly 900 times in 6 months, which is pretty cool to see.</p><p>It was only recently that I took a punt on implementing Stripe to introduce a “Pro” plan. Features aren’t gated, but limits are imposed as they cost money to run. Again, keeping it value-first. The main reason people are there is to browse paint colours, and you don’t need an account to do that.</p><p>What’s next?</p><p>Well, I don’t see Swatcher changing too much. It’ll still be maintained and updated as it has been. Features may be slower to materialise, but only because it already does what it needs to. I could do a better job of surfacing different features, like pointing your camera at something to get a suggested matching paint colour, but I don’t see any massive features coming soon.</p><p>It was architected in a way that the API is independent of the Swatcher front-end. I did this because I was reaching out to paint brands, and could suggest being able to spin up a brand-specific site to showcase their paint colours. I didn’t get any responses, so now I’m thinking about it differently.</p><p>I have a few ideas that I’m working on that can be built from the Swatcher API. I’m pretty excited to explore this, and I’ll post updates if it goes anywhere. Otherwise, you won’t hear from me for a good few months.</p>
      ]]></content>
    </item>
    <item>
      <title>Swatcher Updates</title>
      <link>https://tj.ie/swatcher-updates-2026/</link>
      <description>&lt;p&gt;Since launching Swatcher, many things have changed. The core of it - making paint colours available in one single place - remains the same, and has even expanded with even more paint brands. We’re nearing 8,000 paint colours now.&lt;/p&gt;&lt;p&gt;Everything that matters most is free and doesn’t require an account. For those interested in managing more projects, collecting infinite favourite colours, and using more AI features to generate palettes, there’s now a “pro” plan.&lt;/p&gt;&lt;p&gt;I don’t see it taking off right now, since most people want to preview and find paint colours. I’m going to continue iterating on features and introducing new ones as the year goes on. I’d like to refine the colour capture feature to make it easier to use and keep track of what colours you’ve spotted.&lt;/p&gt;&lt;p&gt;Anyway, I’ll keep an eye on traffic as things pick back up after the Christmas break to help drive some decisions.&lt;/p&gt;&lt;p&gt;Speaking of traffic, here’s how things have gone since July 1st 2025:&lt;/p&gt;&lt;ul&gt;&lt;li&gt;5.3k visitors&lt;/li&gt;&lt;li&gt;14k pageviews&lt;/li&gt;&lt;li&gt;Top Referral: 2.7k from Google&lt;/li&gt;&lt;li&gt;Device Breakdown: 74% mobile, 24% desktop and 0.8% tablet&lt;/li&gt;&lt;li&gt;Top Browsers: 57% Google Chrome, 24% iOS Safari&lt;/li&gt;&lt;li&gt;Top Countries: 61% from Ireland, 23% from the UK and 5% USA&lt;/li&gt;&lt;/ul&gt;&lt;p&gt;I’m working to build on these metrics and improve them this year. Now that I have a solid foundation of basic features and a paid subscription in place, I have more motivation to deliver on value.&lt;/p&gt;&lt;p&gt;&lt;a href=&quot;https://swatcher.ie&quot;&gt;Swatcher&lt;/a&gt;&lt;/p&gt;</description>
      <pubDate>Thu, 15 Jan 26 09:00:00 +0000</pubDate>
      <dc:creator>TJ Fogarty</dc:creator>
      <guid>https://tj.ie/swatcher-updates-2026/</guid>
      <content type="html"><![CDATA[
<p>Since launching Swatcher, many things have changed. The core of it - making paint colours available in one single place - remains the same, and has even expanded with even more paint brands. We’re nearing 8,000 paint colours now.</p><p>Everything that matters most is free and doesn’t require an account. For those interested in managing more projects, collecting infinite favourite colours, and using more AI features to generate palettes, there’s now a “pro” plan.</p><p>I don’t see it taking off right now, since most people want to preview and find paint colours. I’m going to continue iterating on features and introducing new ones as the year goes on. I’d like to refine the colour capture feature to make it easier to use and keep track of what colours you’ve spotted.</p><p>Anyway, I’ll keep an eye on traffic as things pick back up after the Christmas break to help drive some decisions.</p><p>Speaking of traffic, here’s how things have gone since July 1st 2025:</p><ul><li>5.3k visitors</li><li>14k pageviews</li><li>Top Referral: 2.7k from Google</li><li>Device Breakdown: 74% mobile, 24% desktop and 0.8% tablet</li><li>Top Browsers: 57% Google Chrome, 24% iOS Safari</li><li>Top Countries: 61% from Ireland, 23% from the UK and 5% USA</li></ul><p>I’m working to build on these metrics and improve them this year. Now that I have a solid foundation of basic features and a paid subscription in place, I have more motivation to deliver on value.</p><p><a href="https://swatcher.ie">Swatcher</a></p>
      ]]></content>
    </item>
  </channel>
</rss>
